Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!nntp.coast.net!howland.reston.ans.net!EU.net!usenet2.news.uk.psi.net!uknet!usenet1.news.uk.psi.net!uknet!uknet!news.swan.ac.uk!iialan From: iialan@iifeak.swan.ac.uk (Alan Cox) Newsgroups: comp.os.linux.networking,comp.unix.bsd.netbsd.misc,comp.unix.bsd.freebsd.misc Subject: Re: TCP latency Date: 18 Jul 1996 10:09:16 GMT Organization: Institute For Industrial Information Technology Lines: 26 Message-ID: <4sl2gc$jf6@news.swan.ac.uk> References: <31E995B6.573A3461@lambert.org> <4schqh$ssa@panix2.panix.com> <31E9E122.190E8C64@lambert.org> NNTP-Posting-Host: iifeak.swan.ac.uk Xref: euryale.cc.adfa.oz.au comp.os.linux.networking:45654 comp.unix.bsd.netbsd.misc:4107 comp.unix.bsd.freebsd.misc:23903 In article <31E9E122.190E8C64@lambert.org> Terry Lambert <terry@lambert.org> writes: >Using DMA devices for disk and nework I/O, and having a unified >VM cache, the cost will be: > > 1 DMA from disk to memory (you have to do this anyway) > 1 page offset mapping > 1 copy for header data > 1 DMA from memory to network (you have to do this anyway) > >This should save you several copies over FreeBSD, at least one >more copy over Linux (which has user space NFS), and additional No because the Linux NFSD is mmap()ing the pages for the file I/O. You've also ignored a nasty little detail PC network cards don't do for you which is a single pass over memory cost for checksum computation (unless you have something really cute in your cards DMA engine). BTW: On PCI bus you also don't in theory have to do the disk to memory DMA, you can DMA stuff between cards according to the spec. Alan -- Send unsolicited junk mail to this address and maybe win the chance to have yourself added free to several hundred random mailing lists. ,--------------- ------------------------------------------------------------/ Alan Cox This signature comes with a free redistribution license / alan@cymru.net